From generic to specific deep representations for visual recognition | IEEE Conference Publication | IEEE Xplore